Reuse verified Campaign preparation across treatments and startup - #1180
Conversation
|
Important
This repository does not receive automatic reviews because it has fewer than 10 stars. ⚙️ Run configurationConfiguration used: defaults Review profile: CHILL Plan: Advanced Run ID: Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
Codex Review SummaryThis comment shows the latest Codex review activity on this pull request.
ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings. |
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: e28c2c27cd
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
e28c2c2 to
a940519
Compare
Change
Campaign treatments currently rescan the same feature and replay files, and controller startup performs another full freeze. Add a native, versioned
campaign-prepareplan that admits inputs once, renders its members from an immutable snapshot and retains SHA-bound preparation artifacts. Identical plans and partial preparations reuse validated evidence; the canonical controller can start from a prepared freeze after checking its exact source, image, dataset, plan, seed and output bindings.Cache and freeze evidence carry authentication tags anchored in an independently selected existing ACK ledger, including the holdout identity derived at original admission. Caller-supplied hashes alone cannot skip validation. Multi-member plans propagate a shared statistical comparison-family trial bound into both factor screening and the supervised evaluator while each job reserves only its own trials. Native terminal readback rejects a model-scoring policy that omits this bound. Prepared comparisons permit zero automatic follow-ups.
Native authorize/sign/finalize/dispatch, independent worker admission, grants, holdout admission and budget limits remain enforced. Preparation creates only cache-integrity attestations, not execution authorization, grant changes, training or Jobs. Runner/controller images provision the ACK host marker; expanded materialization metadata retains a bounded envelope.
Issue relationship
Closes #1178
Validation
Runtime impact and rollback
No real study or ACK resource was started. Runner/controller images may publish automatically after merge admission. Runtime adoption and research acceptance require their own evidence. Revert this change to remove the new preparation entrypoint and restore full preparation on startup; immutable historical artifacts remain audit records. This preparation slice does not claim the entire workflow repair is finished.